Sleep pressure builds up through a voltage-controlled fat oxidation process

The pore-forming α-subunits of voltage-gated potassium channels of the K_V1 and K_V4 families partner with non-membrane-integral β-subunits^1–5 whose sequences exhibit puzzling similarity with aldo-keto reductases^6,7—enzymes that reduce carbonyls to alcohols via the coupled oxidation of an NADPH cofactor. The isolated β-subunits show weak reductase activity towards a range of model aldehydes in vitro^18,19, relying on NADPH as the electron donor, but whether, on which native carbonyls, and to what end the assembled K_V channel catalyses similar reactions in vivo is unknown. The exceptionally firm grip of K_Vβ on its cofactor⁸, which chokes catalysis, deepens the mystery of why an ion channel would be shackled to what appears to be a subpar enzyme.

A hint at a possible answer has come from studies in Drosophila, where both the K_V1 channel Shaker^20,21 and its β-subunit Hyperkinetic⁷ are needed to sustain normal levels of sleep^22,23. The sleep-regulatory function of the channel complex has been mapped to a small number of sleep-control neurons whose axonal projections target the dorsal fan-shaped body in the central brain^15,17,24 (dFBNs). Sleep need is encoded in the electrical activity of these neurons¹⁶, which fluctuates—in part²⁴—because Hyperkinetic modulates the inactivation kinetics of the Shaker current⁹. During waking, electrons leaking from the saturated transport chains of the inner mitochondrial membrane produce superoxide and other reactive oxygen species (ROS), which convert the K_Vβ pool to the NADP⁺-bound form^9,25. This prolongs the inactivation time constant of the associated potassium conductance^9,18,26,27, strengthens the repolarizing force that restores the resting membrane potential after each spike, and so enables dFBNs to fire at higher rates^9,28.

Although the source (the mitochondrial electron transport chain) and the receiver (Hyperkinetic in complex with Shaker) of the sleep-promoting redox signal are known^9,25, the mode of communication between mitochondria and potassium channels remains undefined. K_Vβ-bound NADPH is an unlikely direct target of ROS, not only because radical-induced hydrogen abstraction (which involves a single electron transfer) will not produce NADP⁺ (which would require the loss of two electrons). As ROS spread from the inner mitochondrial membrane, they encounter many potential reaction partners before reaching Hyperkinetic at the cell surface. Among the most abundant and vulnerable ROS targets in the immediate vicinity of their site of origin are the polyunsaturated fatty acyl chains (PUFAs) of membrane lipids, whose peroxidation and subsequent fragmentation into carbonyls^10–13 can create chemical functionality fit for the active site of an aldo-keto reductase. In the crystal structure of the mammalian K_V1.2–β2 channel complex, the substrate binding pocket is lined with hydrophobic residues and filled with unresolved electron density⁴, as would be expected if a diverse group of lipid precursors disintegrated into a heterogeneous mix of apolar ligands. Recombinant K_Vβ1 and K_Vβ2 reduce synthetic analogues of lipid peroxidation products, such as 4-oxo-2-nonenal (4-ONE), 1-palmitoyl-2-oxovaleroyl-phosphatidylcholine or methylglyoxal, in vitro^18,19, but turnover is so slow that the effect on the concentrations of these molecules in vivo must be minimal. While K_Vβ can therefore have no plausible role in the enzymatic clearance of toxic carbonyls, the very features that seem detrimental or baroque in a catalyst—the protein’s stranglehold on NADP(H) and its linkage to a voltage-gated ion channel—could be essential if the assembly instead functioned as a biochemical memory cell (Fig. 1). Imagine that tight binding of NADP(H) causes the redox reaction to pause at the cofactor-exchange step. Each β-subunit then records a single exposure to an oxidizing substrate by flipping from the NADPH-bound to the NADP⁺-bound form and stores this bit of information until NADP⁺ is released and replaced by NADPH (Fig. 1a). The operational logic resembles that of a single-transistor dynamic random-access memory (DRAM) cell²⁹ (Fig. 1b): K_Vβ corresponds to the storage capacitor of a DRAM cell; the oxidation state of NADP(H) plays the part of the electric charge on the capacitor; and the (low) basal reaction rate is equivalent to the leakage of charge from the capacitor, which gives the memory a finite lifetime that requires periodic refreshment²⁹. The analogy would be complete if, akin to the voltage across the transistor that gates access to the storage capacitor in a DRAM chip²⁹, the membrane potential across the voltage sensors of the α-subunit controlled the rate of cofactor exchange by the β-subunit (Fig. 1).

Here we test several tenets of this model. We examine the lipids of rested and sleep-deprived brains for signs of oxidative damage; measure the effect on sleep of perturbing the clearance of peroxidized lipids; determine whether lipid peroxidation products influence the Shaker current of sleep-control neurons via the active site of Hyperkinetic; and analyse the interplay of voltage sensors and NADP(H) binding sites in the redox regulation of the channel. The results define an autoregulatory loop in which the K_V1 channel population encodes the recent lipid peroxidation history of a neuron in the collective binary states of their β-subunits. This biochemical memory (which we equate to the accumulated sleep pressure) is read and erased during subsequent electrical activity, with the action potential frequency set by the fraction of K_Vβ subunits previously loaded with NADP⁺.

Because levels of oxidative stress may differ among tissues, brain regions or neuron types^9,30, we collected spatial maps of hundreds of lipids by means of high-resolution scanning microprobe matrix-assisted laser desorption/ionization mass spectrometry imaging (SMALDI-MSI). The lipid maps were acquired by scanning 10-µm-thick cryosections of rested or sleep-deprived brains at a lateral resolution of 5 µm × 5 µm and overlaid on fluorescence images of dFBNs expressing R23E10-GAL4-driven¹⁶ mCD8::GFP (Fig. 2a).

Samples within each group had tightly correlated lipid profiles, but differences between groups—that is, between the rested and sleep-deprived states—were so stark that sleep histories could be accurately inferred from lipid composition alone; a single principal component captured 85% of the overall variance. Fifty-one out of 380 SMALDI-MSI signals annotated as glycerophospholipids and detected exclusively on tissue increased or decreased more than twofold after sleep loss, with a false discovery rate (FDR)-adjusted significance threshold of P < 0.05 and little, if any, spatial heterogeneity across the brain (Fig. 2a–c). The identities of 18 of these 51 differentially abundant phospholipids (35%) were confirmed by targeted MS² fragmentation after HPLC separation of a methyl tert-butyl ether extract of brain homogenates (Fig. 2a–c). In many cases these analyses also revealed the detailed fatty acid compositions of the parent species (Fig. 2b,c).

Most lipids with high discriminatory power belonged to one of three classes, which form discernible blocks in the clustergram of Fig. 2b. The glycerophospholipids of rested brains carried inositol, serine, ethanolamine or choline head groups and were enriched in acyl chains with a combined median length of 37.5 carbons and a large degree of unsaturation; the number of double bonds averaged 5.0 ± 2.61 (mean ± s.d.) per lipid, with a median of 5 and a maximum of 12 (Fig. 2b–d). Phospholipids that were present at higher levels in sleep-deprived brains, by contrast, contained mostly choline and ethanolamine head groups, shorter acyl chains with a combined median length of 33.5 carbons, and many fewer double bonds than those in rested flies; the number of double bonds averaged 2.0 ± 2.03 (mean ± s.d.) per lipid, with a median of 2 (Fig. 2b–d). The third distinctive lipid class consisted of several species of phosphatidic acid, whose levels declined after sleep deprivation (Fig. 2b–d). Phosphatidic acid occupies a central position in the biosynthetic pathways of all glycerophospholipids^31,32 and promotes mitochondrial fusion when generated locally by a dedicated phospholipase D (mitoPLD)³³. Impaired mitoPLD activity in dFBNs causes sleep loss²⁵.

The lipidomic fingerprint of sleep-deprived brains indicates that their membranes are depleted of PUFAs, presumably as a consequence of oxidative damage, leaving behind a greater proportion of largely saturated phospholipids (Fig. 2d). The picture during rest is consistent with membrane repair via glycerophospholipid biosynthesis from phosphatidic acid precursors^31,32 and a reversal of the mitochondrial fragmentation that commonly accompanies periods of oxidative stress³⁴, including sleep deprivation²⁵.

The peroxidation of membrane lipids begins^11,12 with the abstraction of a bis-allylic hydrogen from a PUFA chain by a radical oxidant such as HOO• (the conjugate acid of O2−) or •OH. The resulting lipid radical reacts with O₂ to form a lipid peroxyl radical, which propagates the chain by abstracting a hydrogen from another PUFA, generating a new lipid radical and a lipid hydroperoxide^10–13. The reaction continues until two radicals combine in a termination step. The lipid hydroperoxides produced along the way undergo a series of rearrangements and scissions that give rise to a variety of short- and medium-chain carbonyl breakdown products^10–13, including the potential K_Vβ substrate^18,19 4-ONE.

Operating behind a primary bastion of enzymatic and non-enzymatic antioxidants³⁵, soluble short-chain dehydrogenases/reductases, such as carbonyl reductase 1 in mammals^36,37 and its functional homologue sniffer in Drosophila^38,39, form a second defensive ring against lipid peroxidation-derived carbonyls. We examined whether breaching and mending these secondary defences would recapitulate the well-documented effects on sleep of pro- and antioxidant manipulations^9,30,40. Indeed, hemizygous male carriers of the X-linked hypomorphic sniffer allele sni¹ showed increased sleep durations during the day and night (Fig. 3a–c and Extended Data Fig. 1a), owing to vastly extended, hyperconsolidated sleep episodes (Extended Data Fig. 1b,c), at an age before widespread neurodegeneration³⁸ produced locomotor deficits that could have been mistaken for sleep (Extended Data Fig. 1d). Sleep returned to or below wild-type levels when sni¹ mutants expressed a UAS-sni rescue transgene³⁸ (Fig. 3c and Extended Data Fig. 1a), and similarly when the alternative oxidase AOX, which shunts surplus electrons from ubiquinone to H₂O, capped mitochondrial ROS production^9,41 (Fig. 3a,c), or when the putative carbonyl sensor Hyperkinetic was removed by RNA-mediated interference (RNAi), either pan-neuronally or in dFBNs of sni¹ mutant flies (Fig. 3b,c). These data place lipid peroxidation products downstream of mitochondrial respiration in the signalling chain that terminates on the Hyperkinetic pool of dFBNs to raise the pressure to sleep⁹.

To determine whether lipid peroxidation-derived carbonyls could alter the oxidation state of Hyperkinetic’s cofactor, we obtained whole-cell voltage-clamp recordings from dFBNs and estimated the NADP⁺:NADPH ratio of the K_Vβ population from the bi-exponential inactivation kinetics of the A-type current (I_A) (Extended Data Fig. 2): a reduced cofactor increases, whereas an oxidized cofactor decreases, the rate of channel inactivation^9,18,26,27. If PUFA-derived carbonyls are endogenous electron acceptors at the active site of K_Vβ, their ballooning levels in sni¹ mutants^38,39 should drive the Shaker–Hyperkinetic complex into the NADP⁺-bound, slowly inactivating state. Increases in the fast and slow inactivation time constants (τ_fast and τ_slow, respectively) of the A-type current relative to wild-type flies indicate that this was indeed the case (Fig. 4a,b).

Plasma membrane-anchored miniSOG¹⁴ allowed us to switch the cofactor acutely to the oxidized state⁹ and follow its fate thereafter. The exposed chromophore of this light-oxygen-voltage-sensing (LOV) domain protein¹⁴ transfers the energy of blue light efficiently to O₂, producing singlet oxygen (¹O₂) which—presumably indirectly, via a burst of lipid peroxidation—converts the channel population to the NADP⁺-bound form and induces sleep⁹. The oxidation of the cofactor was detected as an increase in the fast and slow inactivation time constants after 9 min of blue light exposure, from initial mean values of 5.8 and 35 ms to final averages of 8.2 and 59 ms (Fig. 4c,d). When the membrane potential was clamped at –80 mV, τ_fast and τ_slow stayed stably elevated for 20 min after the light-driven ¹O₂ generation stopped (Fig. 4c,d), consistent with a negligible rate of spontaneous NADP⁺ exchange^8,18,27 that allows Hyperkinetic to retain a memory of an earlier encounter with an oxidizing substrate, even if that molecule is itself short-lived (estimated intracellular half-life¹⁰ of lipid-derived carbonyls <4 s).

In a direct test of the idea that PUFA-derived carbonyls are prominent among these substrates, we filled dFBNs through the patch pipette with the synthetic lipid peroxidation products 4-ONE or 4-hydroxynonenal (4-HNE)^10,12,13. Owing to their inherent reactivity and membrane-permeability, the equilibration of these carbonyls within the neuronal arbor was governed by complex reaction–diffusion kinetics that made their concentration profiles difficult to predict¹³ and, in all likelihood, neither spatially uniform nor temporally stationary during the course of a recording. 4-ONE and 4-HNE are estimated (with large uncertainty) to be present in cells in the low to sub-micromolar range under basal conditions but reach millimolar concentrations during periods of oxidative stress¹⁰. Although 4-HNE is viewed as a useful marker of lipid peroxidation because monoclonal antibodies can detect its protein adducts⁴², mammalian K_Vβ2 in vitro shows detectable catalytic activity only towards 4-ONE¹⁹. If the substrate preferences of Drosophila Hyperkinetic were similar, 4-HNE could serve as an ideal control to distinguish effects due to the enzymatic conversion of reactive carbonyls from those potentially caused by indiscriminate protein modification¹³.

Comparisons of I_A inactivation kinetics immediately after break-in and 10 min later revealed a clear slowing of the fast and slow time constants, with effect sizes similar to those after the miniSOG-driven photogeneration of ROS (Fig. 4e,f) or a night of mechanical sleep deprivation (Fig. 4g). Changes were seen only in dFBNs perfused with 50 µM 4-ONE; 200 µM 4-HNE, the addition of 0.15% methyl acetate vehicle to the intracellular solution, or the passage of time alone had no effect (Fig. 4g and Extended Data Fig. 3a). When the cells were held at –80 mV in 4-ONE for extended periods, the inactivation time constants completed much of their climbs to higher plateaux within the first 10 min and remained there for the rest of the recordings (Fig. 4f). Because each neuron in this experimental configuration was connected to a practically infinite reservoir of 4-ONE, however, the persistent slowing of inactivation could reflect continuous turnover of substrate rather than a lasting switch in the oxidation state of the cofactor; it can therefore not speak as unequivocally to the longevity of the redox memory as the enduring increase of τ_fast and τ_slow in miniSOG-expressing dFBNs after a finite light exposure can (Fig. 4d).

Membrane resistances, membrane time constants, and the amplitude of the non-A-type potassium current remained approximately constant over the course of 30 min, but the magnitude of I_A slowly declined (Extended Data Figs. 3a,b and 4). This trend is likely to reflect closed-state inactivation⁴³ rather than a gradual loss of voltage control over a portion of the channels before an increase in access resistance would have prompted us to terminate the recording: series resistances stayed within stable limits for 30 min, irrespective of the presence of 4-ONE or changes in command potential or the inactivation kinetics of I_A (Extended Data Fig. 3c), but the steady-state half-inactivation voltages drifted towards more hyperpolarized potentials⁴³ (Extended Data Fig. 3d). Because the same slow rundown of I_A was also observed in the absence of 4-ONE (Extended Data Fig. 3a), after miniSOG stimulation (Extended Data Fig. 4b), and in homozygous Hyperkinetic-null mutants (below), the effect cannot be explained by a direct irreversible 4-ONE hit on the β-subunit.

For the most stringent proof that 4-ONE altered the Shaker current via its reduction at the active site of K_Vβ (as opposed to an off-target modification on the channel or elsewhere), we expressed transgenes encoding catalytically active or dead Hyperkinetic⁴⁴ under R23E10-GAL4 control in dFBNs of Hyperkinetic-null mutant (Hk¹/Hk¹) flies⁹. Infiltrating the Shaker channel with a β-subunit devoid of oxidoreductase activity^18,27,44 (Hk(K289M)) rendered the fast and slow components of A-type inactivation resistant to 4-ONE, whereas the incorporation of functional K_Vβ preserved the sensitivity of the channel (Fig. 4g and Extended Data Fig. 5).

Impaired carbonyl clearance, the photogeneration of ROS, and synthetic 4-ONE exerted indistinguishable effects on I_A in voltage-clamp recordings (Fig. 4a–f), but only carriage of the sni¹ mutation or miniSOG-mediated photooxidation also enhanced the spiking response of dFBNs to membrane depolarization (Fig. 5a,b). The delivery of 4-ONE through a patch electrode at the soma did not (Fig. 5c), in all likelihood because the diffusion time of 4-ONE to dFBN axons, which appear rich in Hyperkinetic but are connected to the cell body through a long, thin primary neurite (Fig. 5d), exceeded the brief intracellular half-life of the molecule^10,13. The release of endogenous lipid peroxidation products, by contrast, whether instigated by miniSOG or amplified by a lack of sniffer, was sufficiently decentralized to be felt also in remote parts of the neuron. The variable spread of externally supplied and internally generated carbonyls will matter little in measurements of voltage-gated potassium currents, which for space-clamp reasons are dominated by channels near the somatic recording site⁴⁵ (Fig. 5d–f), but come to the fore in recordings of action potentials if the spike initiation zone lies outside the diffusion distance of 4-ONE.

The stability of cofactor binding suggests that each conversion of K_Vβ to the NADP⁺-bound state leaves an imprint lasting many minutes (Fig. 4c,d). We equate this imprint—or, more accurately, the imprint on the oxidation state of the Hyperkinetic pool of a dFBN as a whole—with a log of accumulated sleep pressure (Fig. 4g). As in a digital recording, the binary states of many elementary memory cells thus quantize a continuous variable, with a resolution determined by the number of single-bit units. Because sleep pressure is discharged via the electrical activity of dFBNs^16,17, action potentials should erase this memory by releasing NADP⁺ and allowing its replacement with NADPH, whose concentration in the cytoplasm exceeds that of NADP⁺ by at least 40-fold⁴⁶. Such a mechanism would confirm a long-suspected quirk in the enzymatic cycle of K_Vβ and offer a rationale for the association of the protein with a voltage-gated ion channel^8,27.

We tested the prediction that cofactor exchange is voltage-controlled in both of our experimental configurations, using either the photogeneration of ROS by miniSOG (Fig. 6a,b and Extended Data Fig. 6a) or the inclusion of 50 µM 4-ONE in the intracellular solution (Fig. 6c,d and Extended Data Fig. 6b) to load K_Vβ with NADP⁺. Following the expected increases of the fast and slow inactivation time constants at 10 min after break-in, dFBNs were taken through simulated 20-min spike trains at 10 Hz under voltage clamp, with each ‘action potential’ consisting of a 3-ms somatic depolarization to +10 mV. Measurements of τ_fast and τ_slow after this sequence of voltage steps (that is, at 30 min after break-in) showed full reversals of the initial increases driven by miniSOG or 4-ONE (Fig. 6a–d). These reversals were themselves reversible: when dFBNs filled with 4-ONE were held at –80 mV for a further 10 min, the large surplus of 4-ONE in the patch pipette once again drove increases in both inactivation time constants (Fig. 6d), whereas a second 9-min light exposure accomplished the same for miniSOG-expressing cells (Fig. 6b).

Occasionally, the reversal protocol pushed the inactivation time constants below their original baselines, suggesting that depolarization dissipated not only the oxidative strain applied by 4-ONE or miniSOG but also the internally sourced pressure already integrated by the channel complex before the experiment began. Consistent with this idea, dFBNs expressing catalytically inactive^18,27,44 Hk(K289M), which cannot form a redox memory (Fig. 4g and Extended Data Fig. 5a,b), often exhibit the fastest-inactivating A-type currents at baseline⁹ and no modulation by 4-ONE or subsequent voltage changes (Extended Data Fig. 5a,b).

The ability to remember exposures to lipid peroxidation products is an intrinsic property of K_V1 channels, shared by neurons other than dFBNs (Extended Data Fig. 7a–e) and present in mammals, with broad—although not limitless¹⁹—carbonyl selectivity. When HEK-293 cells coexpressing mouse K_V1.4 and K_Vβ2 were incubated in extracellular medium containing 12 mM methylglyoxal, a membrane-permeable dicarbonyl that serves as an established substrate¹⁹ for K_Vβ2, the fast and slow inactivation time constants of the reconstituted A-type current rose and remained durably elevated for 20 min after the removal of methylglyoxal (Extended Data Fig. 7f–h). As in dFBNs, the memory of the carbonyl exposure was retained if the membrane containing the K_V1.4–β2 complex was clamped at −80 mV but forgotten during a simulated 20-min spike train at 10 Hz (Extended Data Fig. 7i,j).

Our experiments suggest that K_Vβ subunits are voltage-gated memories used by neurons and other excitable cells to keep score of lipid peroxidation events. Information is stored in the oxidation state of a nicotinamide molecule bound so tightly that it should perhaps be considered a prosthetic group rather than a cofactor, even though two steps in a stop-and-go redox reaction cycle—hydride transfer and nicotinamide exchange—are used to move data to and from memory. Definitive proof that peroxidized lipids or their breakdown products are endogenous K_Vβ substrates would require their co-purification with the native ion channel—a formidable challenge not only because of the expected molecular heterogeneity of these substrates^10–13, but also because their binding to K_Vβ may be much looser than that of NADP(H); in contrast to the nucleotide binding cleft, which resembles a locked vice, the active site appears wide open in the crystal structure⁸.

Our experiments also suggest, but do not prove beyond doubt, that sleep loss causes widespread lipid peroxidation in the brain. Definitive proof would require a demonstration that peroxidation products accumulate, rather than that polyunsaturated phospholipids are depleted, as we have shown. Most previous attempts to measure lipid peroxidation after sleep loss have focused on a single end product, malondialdehyde¹⁰, and yielded variable results^47–50, perhaps because the picture seen through the lens of malondialdehyde is incomplete⁴² or because the assays used for its detection report tissue oxidizability during analysis rather than pre-existing levels of peroxidized lipids. Our own attempts to quantify endogenous 4-ONE after sleep deprivation were thwarted by the short half-life^10,13 of the molecule in tissue: while SMALDI-MSI could easily detect 100 µM 4-ONE in isolation, the signal vanished when the same quantity of standard was spiked onto a brain section full of endogenous carbonyl-reactive nucleophiles^10,13 and enzymes³⁸ (Extended Data Fig. 8a). Trace amounts of 4-ONE captured by Girard’s reagent during the derivatization of rested, but not sleep-deprived, brains must reflect the oxidation of the undepleted PUFA pools of these samples in vitro because the sni¹ mutation, which would have raised 4-ONE levels in vivo^38,39, caused no discernible increase at the time of measurement (Extended Data Fig. 8b).

While our interpretation of sleep pressure as mitochondrially determined^9,25 lipid peroxidation history demands that sleep-control neurons are equipped to sense and respond to this history, integral redox sensors are a general feature of K_V1 (and also some K_V4) channels^1–5 in virtually all neurons and many other electrically excitable cells. What could be the purpose of β-subunits in this wider context? Redox control of electrical activity may protect non-renewable cells with high respiratory capacity and extensive membrane systems—such as those of the brain and heart—from oxidative damage if the electron supply to their mitochondria surpasses the demands of ATP synthesis²⁵. Depending on where this relief valve opens, the consequences may range from a few extraneous action potentials²⁸ (in order to re-balance energy consumption with mitochondrial electron flux) to the induction of sleep⁹. Just as sodium spikes are universal information carriers filled with distinctive meaning by the different neurons that emit them, excitability control by K_Vβ may be a general mechanism co-opted by dFBNs for the special purpose of regulating sleep.

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-025-08734-4.

The authors thank L. Ballenberger and C. Hartmann for help with dissections and D. Anderson, B. Dickson, B. Ganetzky, T. Holmes, H. Jacobs, J. Ng, G. Rubin, S. Schneuwly, J. Simpson, the Bloomington Stock Center and the Vienna Drosophila Resource Center for flies. This work was supported by grants from the European Research Council (832467) and the UK Medical Research Council (MR/V013238/1) to G.M., and from the German Research Foundation (Sp314/23-1, INST 162/500-1 FUGG) and the Hessian Ministry of Science and Education (LOEWE Center DRUID) to B.S.; H.O.R. and L.G.S. received doctoral training fellowships from Wellcome and La Caixa, respectively; M.A.M. was supported by a Kekulé fellowship from the German Fonds der Chemischen Industrie; P.Z.L. was a Marshall Scholar; and A.K. held postdoctoral fellowships from the Swiss National Science Foundation and EMBO.

H.O.R. performed all electrophysiological and behavioural experiments on flies and M.A.M. performed all lipidomic analyses, under the supervision of S.G. and B.S., on material prepared by L.G.S., H.O.R. and A.K. P.Z.L. characterized Hk^Flag flies and K_V currents in HEK-293 cells. B.S. designed the SMALDI-MSI methodology and instrumentation. G.M. devised and directed the research and wrote the paper.

The SMALDI-MSI and LC-MS² datasets are accessible in METASPACE (https://metaspace2020.eu/project/drosophila↗ and https://metaspace2020.eu/project/drosophila4ONE↗) and the MassIVE repository (ftp://massive.ucsd.edu/v05/MSV000091767/↗), respectively. All other data generated and analysed during this study are included in the Source Data.

M.A.M. and S.G. are employees of and B.S. is a consultant for TransMIT GmbH. The other authors declare no competing interests.

is available for this paper at 10.1038/s41586-025-08734-4.

The online version contains supplementary material available at 10.1038/s41586-025-08734-4.